From: Vinod Koul <vkoul@kernel.org>
To: Tan En De <ende.tan@starfivetech.com>
Cc: dmaengine@vger.kernel.org, Eugeniy.Paltsev@synopsys.com
Subject: Re: [v2,1/1] dmaengine: dw-axi-dmac: Support src_maxburst and dst_maxburst
Date: Wed, 4 Oct 2023 18:59:10 +0530 [thread overview]
Message-ID: <ZR1optUInIcCgTxN@matsya> (raw)
In-Reply-To: <20230913110425.1271-1-ende.tan@starfivetech.com>
On 13-09-23, 19:04, Tan En De wrote:
> Current implementation has hardcoded CHx_CTL.SRC_MSIZE and
> CHx_CTL.DST_MSIZE with a constant, namely DWAXIDMAC_BURST_TRANS_LEN_4.
>
> However, to support hardware with different source/destination burst
> transaction length, the aforementioned values shall be configurable
> based on dma_slave_config set by client driver.
Are client drivers setting this today, will there be regression if not
set?
>
> So, this commit is to allow client driver to configure
> - CHx_CTL.SRC_MSIZE via dma_slave_config.src_maxburst
> - CHx_CTL.DST_MSIZE via dma_slave_config.dst_maxburst
>
> Signed-off-by: Tan En De <ende.tan@starfivetech.com>
> ---
> v1 -> v2:
> - Fixed typo (replaced `return -EINVAL` with `goto err_desc_get` in dma_chan_prep_dma_memcpy()).
> ---
> .../dma/dw-axi-dmac/dw-axi-dmac-platform.c | 38 +++++++++++++++----
> drivers/dma/dw-axi-dmac/dw-axi-dmac.h | 3 +-
> 2 files changed, 33 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c b/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
> index dd02f84e404d..f234097dffb9 100644
> --- a/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
> +++ b/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
> @@ -610,7 +610,7 @@ static int dw_axi_dma_set_hw_desc(struct axi_dma_chan *chan,
> size_t axi_block_ts;
> size_t block_ts;
> u32 ctllo, ctlhi;
> - u32 burst_len;
> + u32 burst_len, src_burst_trans_len, dst_burst_trans_len;
>
> axi_block_ts = chan->chip->dw->hdata->block_size[chan->id];
>
> @@ -674,8 +674,20 @@ static int dw_axi_dma_set_hw_desc(struct axi_dma_chan *chan,
>
> hw_desc->lli->block_ts_lo = cpu_to_le32(block_ts - 1);
>
> - ctllo |= DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_DST_MSIZE_POS |
> - DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_SRC_MSIZE_POS;
> + dst_burst_trans_len = chan->config.dst_maxburst ?
> + __ffs(chan->config.dst_maxburst) - 1 :
> + DWAXIDMAC_BURST_TRANS_LEN_4;
> + if (dst_burst_trans_len > DWAXIDMAC_BURST_TRANS_LEN_MAX)
> + return -EINVAL;
> + ctllo |= dst_burst_trans_len << CH_CTL_L_DST_MSIZE_POS;
> +
> + src_burst_trans_len = chan->config.src_maxburst ?
> + __ffs(chan->config.src_maxburst) - 1 :
> + DWAXIDMAC_BURST_TRANS_LEN_4;
> + if (src_burst_trans_len > DWAXIDMAC_BURST_TRANS_LEN_MAX)
> + return -EINVAL;
> + ctllo |= src_burst_trans_len << CH_CTL_L_SRC_MSIZE_POS;
> +
> hw_desc->lli->ctl_lo = cpu_to_le32(ctllo);
>
> set_desc_src_master(hw_desc);
> @@ -878,7 +890,7 @@ dma_chan_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst_adr,
> size_t block_ts, max_block_ts, xfer_len;
> struct axi_dma_hw_desc *hw_desc = NULL;
> struct axi_dma_desc *desc = NULL;
> - u32 xfer_width, reg, num;
> + u32 xfer_width, reg, num, src_burst_trans_len, dst_burst_trans_len;
> u64 llp = 0;
> u8 lms = 0; /* Select AXI0 master for LLI fetching */
>
> @@ -936,9 +948,21 @@ dma_chan_prep_dma_memcpy(struct dma_chan *dchan, dma_addr_t dst_adr,
> }
> hw_desc->lli->ctl_hi = cpu_to_le32(reg);
>
> - reg = (DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_DST_MSIZE_POS |
> - DWAXIDMAC_BURST_TRANS_LEN_4 << CH_CTL_L_SRC_MSIZE_POS |
> - xfer_width << CH_CTL_L_DST_WIDTH_POS |
> + dst_burst_trans_len = chan->config.dst_maxburst ?
> + __ffs(chan->config.dst_maxburst) - 1 :
> + DWAXIDMAC_BURST_TRANS_LEN_4;
> + if (dst_burst_trans_len > DWAXIDMAC_BURST_TRANS_LEN_MAX)
> + goto err_desc_get;
> + reg |= dst_burst_trans_len << CH_CTL_L_DST_MSIZE_POS;
> +
> + src_burst_trans_len = chan->config.src_maxburst ?
> + __ffs(chan->config.src_maxburst) - 1 :
> + DWAXIDMAC_BURST_TRANS_LEN_4;
> + if (src_burst_trans_len > DWAXIDMAC_BURST_TRANS_LEN_MAX)
> + goto err_desc_get;
> + reg |= src_burst_trans_len << CH_CTL_L_SRC_MSIZE_POS;
this is wrong, mempcy should never use slave config value. These values
are for peripheral and not meant for mem-mem transfers
You should try max throughput by aligning the txn to bus width available
on this system
> +
> + reg = (xfer_width << CH_CTL_L_DST_WIDTH_POS |
> xfer_width << CH_CTL_L_SRC_WIDTH_POS |
> DWAXIDMAC_CH_CTL_L_INC << CH_CTL_L_DST_INC_POS |
> DWAXIDMAC_CH_CTL_L_INC << CH_CTL_L_SRC_INC_POS);
> diff --git a/drivers/dma/dw-axi-dmac/dw-axi-dmac.h b/drivers/dma/dw-axi-dmac/dw-axi-dmac.h
> index eb267cb24f67..877bff395740 100644
> --- a/drivers/dma/dw-axi-dmac/dw-axi-dmac.h
> +++ b/drivers/dma/dw-axi-dmac/dw-axi-dmac.h
> @@ -265,7 +265,8 @@ enum {
> DWAXIDMAC_BURST_TRANS_LEN_128,
> DWAXIDMAC_BURST_TRANS_LEN_256,
> DWAXIDMAC_BURST_TRANS_LEN_512,
> - DWAXIDMAC_BURST_TRANS_LEN_1024
> + DWAXIDMAC_BURST_TRANS_LEN_1024,
> + DWAXIDMAC_BURST_TRANS_LEN_MAX = DWAXIDMAC_BURST_TRANS_LEN_1024
> };
>
> #define CH_CTL_L_DST_WIDTH_POS 11
>
> base-commit: 3669558bdf354cd352be955ef2764cde6a9bf5ec
> --
> 2.34.1
--
~Vinod
next prev parent reply other threads:[~2023-10-04 13:29 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-09-13 11:04 [v2,1/1] dmaengine: dw-axi-dmac: Support src_maxburst and dst_maxburst Tan En De
2023-10-04 13:29 ` Vinod Koul [this message]
2023-11-06 8:44 ` EnDe Tan
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ZR1optUInIcCgTxN@matsya \
--to=vkoul@kernel.org \
--cc=Eugeniy.Paltsev@synopsys.com \
--cc=dmaengine@vger.kernel.org \
--cc=ende.tan@starfivetech.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox